PURPOSE We assessed the long-term outcome of patients enrolled on CCG-3891, a high-risk neuroblastoma study in which patients were randomly assigned to undergo autologous purged bone marrow transplantation (ABMT) or to receive chemotherapy, and subsequent treatment with 13-cis-retinoic acid (cis-RA). PATIENTS AND METHODS Patients received the same induction chemotherapy, with random assignment (N = 379) to consolidation with myeloablative chemotherapy, total-body irradiation, and ABMT versus three cycles of intensive chemotherapy. Patients who completed consolidation without disease progression were randomly assigned to receive no further therapy or cis-RA for 6 months. Results The event-free survival (EFS) for patients randomly assigned to ABMT was significantly higher than those randomly assigned to chemotherapy; the 5-year EFS (mean +/- SE) was 30% +/- 4% versus 19% +/- 3%, respectively (P = .04). The 5-year EFS (42% +/- 5% v 31% +/- 5%) from the time of second random assignment was higher for cis-RA than for no further therapy, though it was not significant. Overall survival (OS) was significantly higher for each random assignment by a test of the log(-log(.)) transformation of the survival estimates at 5 years (P < .01). The 5-year OS from the second random assignment of patients who underwent both random assignments and who were assigned to ABMT/cis-RA was 59% +/- 8%; for ABMT/no cis-RA, it was 41% +/- 7%; for continuing chemotherapy/cis-RA, it was 38% +/- 7%; and for chemotherapy/no cis-RA, it was 36% +/- 7%. CONCLUSION Myeloablative therapy and autologous hematopoietic cell rescue result in significantly better 5-year EFS and OS than nonmyeloablative chemotherapy; cis-RA given after consolidation independently results in significantly improved OS.